Improving Aerodynamic Matching

of Axial Compressor Blading

Using a Three-Dimensional
Multistage Inverse Design
Method
M. P. C. van Rooij1 Current turbomachinery design systems increasingly rely on multistage CFD as a means
to diagnose designs and assess performance potential. However, design weaknesses at-
T. Q. Dang tributed to improper stage matching are addressed using often ineffective strategies in-
volving a costly iterative loop between blading modification, revision of design intent,
Syracuse University, and further evaluation of aerodynamic performance. A scheme is proposed herein which
Syracuse, NY 13244 greatly simplifies the design point blade row matching process. It is based on a three-
dimensional viscous inverse method that has been extended to allow blading analysis and
design in a multi-blade row environment. For computational expediency, blade row cou-
L. M. Larosiliere2 pling is achieved through an averaging-plane approximation. To limit computational
U.S. Army Research Laboratory, time, the inverse method was parallelized. The proposed method allows improvement of
NASA Glenn Research Center, design point blade row matching by direct regulation of the circulation capacity of the
Cleveland, OH 44135 blading within a multistage environment. During the design calculation, blade shapes are
adjusted to account for inflow and outflow conditions while producing a prescribed
pressure loading. Thus, it is computationally ensured that the intended pressure-loading
distribution is consistent with the derived blading geometry operating in a multiblade
row environment that accounts for certain blade row interactions. The viability of the
method is demonstrated in design exercises involving the rotors of a 2.5 stage, highly
loaded compressor. Individually redesigned rotors display mismatching when run in the
2.5 stage, evident as a deviation from design intent. However, simultaneous redesign of
the rotors in their multistage environment produces the design intent, indicating that
aerodynamic matching has been achieved. 关DOI: 10.1115/1.2372773兴

Keywords: inverse aerodynamic shape design, multistage turbomachinery CFD, com-

pressor stage matching

Introduction row interactions that would lead to deviations from this design
intent. The “design intent” implies a physically realizable
Turbomachinery computational fluid dynamics 共CFD兲, in sup-
axisymmetric-averaged pressure field that is compatible with
port of aerodynamic design, has evolved from isolated blade row
overall design point performance and off-design operability. For a
methods to hierarchical multistage analysis encompassing various
given overall pressure rise, mismatching is manifested as a dis-
blade row coupling schemes. Three main approaches to blade row
ruption of the design equilibrium among the circulation capacity
coupling exist: steady averaging plane, time mean average pas-
of the blading, entropy production, and accumulation of aerody-
sage, and unsteady time periodic. The foundations of these ap-
namic blockage. Naturally, an appropriate choice of design intent
proaches are discussed in Refs. 关1,2兴. Currently, multistage CFD
can limit strong blade row interactions.
methods are primarily suited for diagnosing design shortcomings
There are two key elements in the blade row matching problem.
originating from strong blade row interactions that adversely im-
First, there is the establishment of accurate and physically realiz-
pact performance and operability. Typically, simulation results are
able design intent and, second, when necessary, an efficient means
used to revise aerodynamic matching conditions for individual
for tailoring blade shapes so that design equilibrium is restored
blade rows and stages. The term aerodynamic matching is loosely
among blading circulation capacity, entropy production, and accu-
understood to involve the compatibility of the inlet flow require-
mulation of aerodynamic blockage. Transforming overall func-
ments of a stage to the outlet flow of upstream stages.
tional requirements into credible aerodynamic design intent is an
In the title of this paper, the word “matching” denotes a critical
inherently iterative process guided by past experience. Note that
function of the multi-blade row design process, where the indi-
this is compatible with the traditional model for turbomachinery
vidual blade row shapes are simultaneously tailored in such a way
aerodynamic design, as described by Marble 关3兴, whereby the
as to produce a desired design intent while accounting for blade
flow is conceptually decomposed into two parts: that which is due
to the gross influence of all blade rows and that which is associ-
1
Present address: Siemens Power Generation Industrial Applications. ated with local details of blade shape.
2
Present address: Concepts NREC. A contemporary aerodynamic design system, incorporating
Contributed by the International Gas Turbine Institute 共IGTI兲 of ASME for pub- multistage CFD analysis, is noted in Fig. 1共a兲. This system allows
lication in the JOURNAL OF TURBOMACHINERY. Manuscript received October 1, 2004;
final manuscript received February 1, 2005. IGTI Review Chair K. C. Hall. Paper
two broad functions: design synthesis, and design diagnostics and
presented at the ASME Turbo Expo 2005: Land, Sea and Air, Reno, NV, June 6–9, development. Both of these functions can be iteratively executed
2005, Paper No. GT2005-68271. to arrive at credible design intent and a blading revision strategy

108 / Vol. 129, JANUARY 2007 Copyright © 2007 by ASME Transactions of the ASME

Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 1 „a… Current design system and „b… proposed design system incorporating 3D inverse blading design
to meet this intent. For advanced designs, this system can become
very costly and ineffective due to the need for many revisions that
are sometimes formulated in an ad-hoc fashion. Thus, enhance-
ments to this process would be beneficial in terms of reductions in
design cycle cost and time. Figure 1共b兲 shows a proposed en-
hancement to the contemporary aerodesign system. This enhance-
ment is made possible by the following two developments: 共1兲
incorporation of an inverse blade-design procedure into the mul-
tistage CFD, and 共2兲 improved consistency between the passage-
averaged throughflow model, the multistage CFD analysis, and
inverse blade design. Note that the enhanced design system in-
volves automatic exchange of appropriate information between
3D viscous multistage analysis, inverse blade design, and passage-
averaged throughflow. This is achieved by a so called averaged-
flow closure and pressure-loading manager which regulates the
design intent, thereby reducing and perhaps eliminating the need
for circuitous and possibly ad-hoc design revision strategies. Work
on the pressure-loading manager, possibly involving neural net-
works, is in its infancy and will be forthcoming.
In this paper, a three-dimensional viscous inverse blade design
method with multirow analysis and indirect geometric sculpting
capabilities is presented. Currently, blade row coupling is done via
an averaging plane, mainly for computational expediency. Thus,
the major axisymmetric blade row interactions are captured while
the nonaxisymmetric coupling effects involving deterministic un-
steadiness are neglected. Debate still continues as to the impact of
the unsteady deterministic flow on time-mean performance, espe-
cially near design point conditions. Does one need to accurately
predict the consequences of mismatching in order to diagnose the
existence of improperly matched blade rows? No attempt is made
to resolve this issue here; instead, ultimate model selection is left
up to the designer. Suffice it to say that the current inverse design
method can be incorporated into higher-fidelity blade row cou-
pling schemes. The usefulness of this approach for improving de-
sign point blade row matching by direct regulation of the circula-
tion capacity of the blading within a multistage environment is
illustrated via a redesign of supersonic rotors for a highly loaded
2.5-stage compressor.
Development of Methodology
The goal of blading design is to effectively realize an intended
velocity diagram with minimal loss and the widest possible oper-
ating range. This is achieved by tailoring the blade pressure load-
ing distribution as is reflected in the angular momentum-force
balance relationship for a quasi-3D blade element
Journal of Turbomachinery
Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
冕
TE
r⌬pdA␪ ⬇ ṁ关共rV̄␪兲TE − 共rV̄␪兲LE兴
LE
In the above relation, ⌬p 共i.e., blade pressure loading兲 is the dif-
ference between the blade upper and lower surface pressures at
fixed axial locations; the subscripts LE and TE denote leading and
trailing edges; ṁ is the mass flow rate; A␪ is the projected surface
area in the tangential direction; and rV̄␪ is the mass-averaged swirl
velocity. Thus, the overall blading design strategy can be de-
scribed as pressure-loading tailoring to attain local aerodynamic
control while satisfying certain global constraints such as net cir-
culation, mass flow rate, and blade count.
The current multistage inverse design method is an extension of
the code INV3D, which has been reported in previous papers by
Dang 关4兴, Qiu 关5兴, Damle et al. 关6兴, Qiu and Dang 关7兴, Dang et al.
关8兴, and Medd 关9兴. Note that INV3D can be used for both design
and analysis. In the inverse mode, the primary prescribed quanti-
ties are the blade axis definition in terms of its axial grid-line
location iba and tangential orientation along the camber surface
f ba共r兲, the blade thickness distribution T共r , z兲, and the pressure
loading distribution ⌬p共r , z兲. Note that the blade axis gives a ref-
erence for locating the various spanwise blade sections in space
and thus must be compatible with the spanwise distribution of ⌬p.
For a given set of inputs, the 3D inverse method computes the
corresponding wrap angle f共r , z兲. Thus, this is a semi-inverse pro-
cedure in the sense that the full geometry is not evolved and only
the mean camber surface, f共r , z兲, is updated via the flow-tangency
condition along the blade surfaces. Clearly, the blade geometry
corresponding to prescribed values for 关iba , f ba , T , ⌬p兴 and operat-
ing in a particular inflow/outflow environment may not satisfy
certain geometric smoothness criteria and is not guaranteed to
have optimum performance nor be aeromechanically acceptable.
The challenge is to pick these quantities to arrive at a satisfactory
design.
The computational method is based on the solution of the three-
dimensional Reynolds-averaged Navier–Stokes 共RANS兲 equations
on a simple sheared H grid using the robust finite-volume time-
marching cell-centered scheme of Jameson et al. 关10兴. All bound-
ary layers are assumed turbulent and viscous effects are modeled
using wall functions with an adaptation of the Baldwin–Lomax
turbulence closure. Normally, INV3D employs a “slip” velocity on
the blade surface, since the inverse scheme requires a finite veloc-
ity to trace the surface 共tangency condition兲, as discussed earlier.
In this case, the viscous sublayer is neglected subject to the con-
dition that the first cell next to a solid boundary lies in the loga-
JANUARY 2007, Vol. 129 / 109
 Fig. 2 Comparison between measured data and CFD simulations of NASA rotor 37 using INV3D
and ADPAC: „a… performance map of total pressure ratio versus mass flow and „b… exit spanwise
profile of total-to-total adiabatic efficiency at nominal operation
rithmic region and hence the existence of a slip velocity 关11兴.
Typical viscous calculations using INV3D employ near-wall y+ on
the order of 30–150. Leakage flows are approximated by assum-
ing periodicity across the extended blade surfaces in the clearance
gap.
In order to transform the original research code into a practical
tool, several modifications were made to the basic methodology
described by Dang 关4兴 and Dang et al. 关8兴. The two most critical
modifications are: 共1兲 the solution method for the camber surface
generator, and 共2兲 a hybrid inverse/analysis scheme to handle the
blade leading and trailing edges.
In the original method, the camber surface is developed by
enforcing the condition of zero-normal velocity on the blade sur-
face. The resulting equation for the camber wrap angle f共r , z兲 is a
convective equation, the so-called camber generator equation. It is
basically a stream surface tracer, the stream surface being the
blade surface. In the original method, the camber generator equa-
tion is solved using the implicit Crank–Nicholson scheme. This
approach was demonstrated to work very well for inviscid flows
关8兴, but it is not robust for practical problems involving significant
viscous effects and leakage flows. In the latter case, the streamline
in the clearance region is highly three dimensional, resulting in
the blade shape being exceedingly distorted during the iterative
process.
Rather than solving the camber generator “exactly” with a con-
ventional implicit numerical method as in the original methodol-
ogy described in Dang et al. 关8兴, the camber generator equation is
satisfied by minimization in the least squares sense, with the span-
wise variation of the camber defined in terms of a NURBS curve
关9兴. In essence, the camber is generated by fitting the blade as best
as possible to the velocity field. This is done in the spanwise
direction at each axial location of the grid, followed by a geo-
metrical smoothing in the chordwise direction. The originally
specified blade axis is maintained. This method is very robust, and
makes it possible to obtain smooth blade geometries and good
convergence in the presence of strong clearance leakage flow.
As this is a minimization process, it cannot be guaranteed that
the local normal velocity be driven to zero during convergence of
the design calculation. This again implies that the specified load-
ing distribution may not be strictly enforced. However, this is
outweighed by the advantages of increased robustness and geo-
metrically smoother blade shapes. To obtain a satisfactory balance
between smoothness and accuracy, the number of NURBS control
points used in the spanwise and chordwise directions can be var-
ied. The more points used, the closer the normal velocity is driven
to zero and the more strictly the specified loading is enforced, but
at a cost of smoothness and possibly convergence 共e.g., in the
presence of strong leakage flow兲.
110 / Vol. 129, JANUARY 2007
Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Another important modification to the original method is the
implementation of a hybrid inverse/analysis technique to handle
the blade leading and trailing edges. Experience with application
of the inverse method to the design of highly loaded transonic
blades, revealed the need for the inverse method to preserve the
blade leading-edge detailed geometry in order to capture impor-
tant local flow structures. For example, the grid must be clustered
near the leading edge so that the incoming flow sees the blade
leading edge as blunt rather than sharp. However, highly clustered
sheared H grids near the leading edge can result in convergence
difficulties initiated by possible distortion of the blade leading
edge shape. This is due to the fact that near the leading edge, grid
skewness, and dispersion errors can create a locally distorted ve-
locity field, and since the blade shape is traced from this velocity
field, it too will be distorted. Thus, in order to maintain geometri-
cally accurate leading and trailing edge shapes during the design
process, a hybrid inverse technique is used 关9兴. Here, the leading
and trailing edge geometries are created by extrapolation of the
blade camber surface from the interior using a NURBS curve. The
meridional envelope of the blade is maintained. Analysis bound-
ary conditions are specified in these regions, which typically en-
compass 1–5% of the blade chord. Apart from geometric fidelity,
this method also increases robustness of the camber generator in
the presence of exceedingly high swirl angles that are typical of
multistage end wall flows at the design point. It may not be de-
sirable to sustain high local gradients in the blade edge angles, and
the 3D relief phenomenon discussed in Wadia and Beacher 关12兴
tends to accommodate smooth blades.
The INV3D solver is able to predict the static pressure field with
reasonable accuracy along with adequate resolution of most criti-
cal flow structures responsible for relative changes in the entropy
production mechanisms 共e.g., when massive flow separation is not
present兲. Figure 2 shows overall total pressure ratio characteristics
and axisymmetric-averaged spanwise profiles of adiabatic effi-
ciency for NASA rotor 37 关13兴 comparing results from INV3D both
with and without wall slip against measurements 关14兴. Also indi-
cated, are results using the NASA developed multiblock RANS
code, ADPAC 关15兴. Both INV3D and ADPAC show good agreement
with the spanwise distribution of efficiency inferred from mea-
surements. The absolute level of total pressure is not well pre-
dicted and could be improved with better viscous and/or turbu-
lence modeling. Overall, the agreement with measurements is
adequate in the sense that INV3D does identify critical flow struc-
tures responsible for relative changes in performance.
INV3D was extended to multistage by adding a blade row cou-
pling scheme and parallelization. Currently, blade row coupling
for multistage calculations is achieved by a steady averaging-
plane approximation. Two versions have been implemented. The
Transactions of the ASME

Fig. 3 Meridional flow path for the 2.5-stage compressor used
in the design examples: Inlet guide vane „IGV…, rotor 1 „R1…,
stator 1„S1…, rotor 2 „R2… and stator 2 „S2…; rotor 1: 26 blades,
average solidityⴝ2.16, aspect ratioⴝ0.83; rotor 2: 56 blades,
average solidityⴝ1.96, aspect ratioⴝ0.86. Also indicated, are
multistage computational averaging planes.
first involves direct exchange of circumferentially averaged den-
sity, momentum and pressure across the averaging-plane 关11,16兴.
The second, considered to be numerically superior, is an adapta-
tion of the nonreflecting averaging plane presented by Chima 关17兴,
which is based on characteristic boundary conditions derived by
Giles 关18兴. It is recognized that the steady averaging-plane cou-
pling ignores much, if not all, of the physics associated with de-
terministic unsteadiness inherent to multistage turbomachinery
flows. However, it does provide spanwise consistency in the con-
servative variables along with global mass conservation thereby
allowing first-order axisymmetric blade row interactions to be
captured. Due to ease of implementation and computational con-
venience, as a first step, the averaging-plane blade row coupling
was therefore chosen for exploring the potential of blade inverse
design in a multistage environment.
Because of the extensive computational demand of a multistage
calculation, the ability to exploit parallel processing computing
capability has been added to the code in the form of MPICH
关19,20兴. For a parallel run, each blade row is assigned to a slave
process on a separate processor, while a master process handles all
the input and output. A cluster of 5 AMD XP2400⫹ PCs running
Windows 2000 has been used for all calculations presented here.
Aerodynamic Matching Study
A meridional cross section of the 2.5-stage advanced compres-
sor used to illustrate the viability of blade inverse design for im-
proving blade row matching in a multistage environment is shown
in Fig. 3. The design philosophy of this compressor is given in
Larosiliere et al. 关21兴; however, the current blading set is a variant
of that described in the reference. Design features include variable
inlet guide vane, cantilevered stators, and 3D blading. The com-
pressor can be characterized as generally high Mach number, high
reaction, and highly loaded. Rotor 2 has supersonic relative inlet
Mach numbers along its entire span, thus exacerbating the stage-
matching problem. The aerodynamic design requirements for this
compressor are a corrected mass flow of 31.8 kg/ s 共70 lbm/ s兲, an
overall total compression ratio of 4.65:1 共at shroud backpressure
of 4.0兲, a total pressure ratio of 2.42:1 across rotor 1 with a cor-
rected tip speed of 450.2 m / s 共1477 ft/ s兲, and a rotor 2 total
pressure ratio of 2.0:1 at a corrected tip speed of 393.2 m / s
共1290 ft/ s兲.
This compressor was originally designed using a contemporary
design system similar to that noted in Fig. 1共a兲. An earlier isolated
blade row version of INV3D 共i.e., rudimentary viscous model, low
mesh resolution, and basic inverse scheme兲, with aerodynamic
matching conditions extracted from a design point throughflow
model, was used to refine the rotor blade shapes in an attempt to
better manage the passage shock structure and strength 共see Ref.
关21兴 for details兲. Note that the credibility of the design-intent
axisymmetric-averaged pressure field from the design point
Journal of Turbomachinery
Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
throughflow model is debatable and indicates the need for estab-
lishing consistency among the various mathematical formulations
and physical model closures. Overall, the blading design objective
was to achieve efficient design point operation at very high aero-
dynamic loading levels 共i.e., average Lieblein D-factors ⬃0.55兲.
The computational mesh used in the present study consisted of
51 cells in the axial direction along the blade surface 共for each
blade row兲, 45 cells in the radial direction, and 35 cells in the
pitchwise direction. Three cells were placed in the rotor tip and
stator hub gaps 共clearances ranging from 0.5% to 1% local chord兲.
In the analysis mode, this mesh size requires approximately 1 h
for 1500 time steps using 5 AMD XP2400⫹ processors 共one pro-
cessor for each blade row兲. A single blade row calculation takes
about 2000 iterations to convergence. A multistage calculation
typically requires more iterations, since errors propagate through
multiple blade rows and the blade row coupling itself affects over-
all convergence. For the 2.5-stage geometry presented here, about
6000 iterations were needed to reach convergence, corresponding
to a computational time of 4 h. Compared to the analysis mode,
the inverse mode takes about 10% longer per iteration. For all
cases presented, the same profiles of total pressure, total tempera-
ture, and radial and absolute tangential flow angles were pre-
scribed at the inlet of the computational domain. A design point
backpressure of 4.00 共normalized to inlet total pressure at mid-
span兲 was specified at the casing, and simple radial equilibrium
was used to construct the spanwise exit static pressure profile. The
standard averaging-plane approach 关11,16兴 was used for steady
blade row coupling.
Diagnosis of Original Blading Geometry. The complete 2.5
stage with the original blading geometry was analyzed at the
design-intent backpressure and speed using INV3D in the analysis
mode. Diagnostic plots of results for the original geometry are
shown in Figs. 4–6, which are described in the following. Figure
4 shows contours of relative Mach number at 3%, 50%, and 100%
span. Three important conclusions can be made by inspection: 共1兲
generation of significant aero blockage by strong leakage/shock
interaction near the casing end wall of both rotors, 共2兲 existence of
a strong quasi-normal passage shock at the mouth of rotor 2 ex-
tending from hub to near midspan, and 共3兲 evolution of separated
suction-side corner flow near rotor 2 hub inducing increased aero
blockage production in the downstream stator 共S2兲 hub clearance
region. Thus, based on INV3D analysis at the design-intent back-
pressure, the original blading set is forced into an off-design op-
eration due to aerodynamic mismatching.
Further insight into the nature of the design point mismatching
can be gleaned from the predicted pressure-loading distributions
for R1 共Fig. 5兲 and R2 共Fig. 6兲. These figures show three plots
each. The first plot indicates the pressure-loading distribution,
⌬p共r , z兲, for the original blading resulting from a multistage
analysis at the design backpressure; the second plot is for an iso-
lated analysis of each rotor using inflow/outflow conditions from
the design point throughflow model; and the third plot is the
design-intent pressure-loading distribution to be discussed shortly.
A comparison of the multistage and isolated analysis results given
in Fig. 5 indicates that R1 does not experience significant multi-
stage design point mismatching effects; rather there exists a strong
tip leakage/shock interaction resulting in larger aero blockage than
intended. A similar comparison for R2, shown in Fig. 6, indicates
significant multistage design point mismatch. The hub corner
separation appears to be eliminated or greatly reduced for the
isolated analysis of R2, which can be discerned by a comparison
of the two loading distributions 共i.e., Fig. 6共a兲 versus Fig. 6共b兲
with low ⌬p implying separation兲 along the hub section. It is
evident that the passage shock for R2 is close to being spilled.
Indeed, throttling the 2.5 stage to a higher backpressure of 4.02
leads to the passage shock of R2 being spilled completely, shortly
followed by a numerical stall of the 2.5 stage.
To explore the viability of the inverse method for facilitating
JANUARY 2007, Vol. 129 / 111
 Fig. 4 Diagnosis „multistage analysis… of flow structure induced by original geometry at de-
sign point in terms of contours of relative Mach number at 3%, 50%, and 100% „rotor blade tip…
span. IGV and exit region are not shown for clarity.

and improving design point matching, two design exercises have throughflow model of the original design. The second involves the
been executed. In the first, R1 and R2 are redesigned in isolation redesign of both rotors simultaneously within the multistage en-
using the latest improved version of INV3D. Inflow and outflow vironment. Both cases employ exactly the same pressure-loading
conditions are specified in accordance with a design point distributions. Design-intent pressure-loading distributions for the

112 / Vol. 129, JANUARY 2007 Transactions of the ASME

Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 5 Predicted pressure-loading distribution for original R1
at design point: „a… multistage analysis of original design, „b…
isolated analysis of original R1 with throughflow conditions,
and „c… prescribed design intent for redesign exercise
redesigns are shown in Figs. 5 and 6, respectively, for R1 and R2.
The objective is for both rotors to have a single swallowed pas-
sage shock, positioned such that the shock extends from near the
trailing edge at the blade tip to the leading edge at around 25%
span, where the shock weakens as it enters a region of low-
supersonic or subsonic flow at lower spanwise positions. These
loading distributions, derived from numerical experiments, are
chosen for the purpose of achieving reduced tip leakage/shock
interaction, and to enable larger design-speed throttling range. A
detailed explanation of the rationale behind these loading distri-
butions is given in Refs. 关9,22兴. Furthermore, the separation re-
gion near the hub of R2 is to be removed, which should not only
improve overall throttling capability but also provide better
matching with Stator 2.
For the design-intent pressure-loading distribution of R1, the
spanwise loading area distribution was taken directly from the
analysis result of the original 2.5 stage at design backpressure. For
R2, the spanwise distribution of loading area was altered slightly,
increasing the loading near the hub and decreasing it elsewhere,
while maintaining the total loading volume and therefore work
done by the rotor. This was done to counter the separation region
close to the hub. The loading area distributions used here are not
directly tied to any “optimized” design-intent pressure field, espe-
cially considering that the passage shock for the original R2 is
Fig. 6 Predicted pressure-loading distribution for original R2
at design point: „a… multistage analysis of original design, „b…
isolated analysis of original R2 with throughflow conditions,
and „c… prescribed design intent for redesign exercise
Journal of Turbomachinery
Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 7 Isolated analysis of R1 and R2 „case A… at design point
in terms of midspan contours of relative Mach number
nearly spilled. However, the aim here is not to obtain a “best”
axisymmetric-averaged design-intent pressure field, but to demon-
strate the feasibility of the multistage design method to enforce a
prescribed design-intent pressure-loading distribution and thus fa-
cilitate design point blade row matching. Note that the proposed
pressure-loading manager indicated in Fig. 1共b兲 would attempt to
manipulate the loading area in order to achieve design-intent av-
eraged static pressure distributions consistent with other pre-
scribed quantities. For both rotors, the axial position of the blade
axis was set at about 20% axial chord.
Design Case A: Rotors 1 and 2 Redesigned in Isolation. Both
rotors were redesigned in isolation. Axisymmetric-averaged pro-
files of total pressure, total temperature and flow angles, obtained
from the design point throughflow model, were specified as inflow
conditions. Backpressures, also obtained from the throughflow
model, of 1.645 for R1 and 3.48 for R2 were specified at the
shroud, with radial equilibrium used to compute the spanwise dis-
tribution of static pressure. The newly obtained geometries, de-
noted as R1A and R2A, were then analyzed in isolation, with the
same “throughflow” boundary conditions used for the redesign.
Predicted relative Mach number distributions at midspan are
shown in Fig. 7 for later comparison.
Figure 8 gives a design point 共pb = 4.00兲 diagnosis of the flow
structure as predicted by a multistage analysis of the 2.5 stage
incorporating the newly designed R1A and R2A. For both rotors,
the design-intent pressure-loading distribution along with its im-
plied shock structure is nearly realized. However, closer inspec-
tion of Figs. 7 and 8 reveal the existence of aerodynamic mis-
matching manifested by the relative changes in shock structure
between isolated and multistage operations. The passage shock for
R2A 共Fig. 8兲 is situated further upstream in the multistage envi-
ronment than intended 共Fig. 7兲. Similarly, R1A is throttled lower
than intended in the multistage environment. Both rotors exhibit
reduced tip leakage/shock interaction resulting in a much cleaner
casing endwall flow than that of the original design. The suction-
side hub corner separation is subdued and a significantly reduced
level of aero blockage is generated in the hub clearance region of
Stator 2 共Fig. 4 versus Fig. 8兲.
The changes in flow field 共and therefore pressure-loading兲 from
“intent” imply some sort of mismatching of the rotors to their
surroundings. Noting the changes in Mach number levels that
have occurred within the 2.5 stage, this in itself is not surprising.
It does raise many questions, one of which is whether the ob-
served changes could have been predicted and accounted for in
the redesign. A simpler question to be answered, which is at-
tempted here, is the following: if a blade is redesigned within its
multistage environment, will the design adapt to this 共possibly
JANUARY 2007, Vol. 129 / 113
 Fig. 8 Diagnosis of flow structure induced by design case A geometry at design point in terms
of contours of relative Mach number at 3%, 50%, and 100% „rotor blade tip… span. IGV and exit
region are not shown for clarity.

changing兲 environment in order to produce the desired intent in Here, both rotor 1 and rotor 2 were redesigned simultaneously
the form of specified pressure-loading distribution? To this end, within the 2.5 stage. The redesign calculation was restarted from
design case B was performed. the multistage analysis calculation of R1A and R2A. The exact
Design Case B: Rotors 1 and 2 Redesigned in 2.5 stage. same loading shapes and loading area distributions were used as

114 / Vol. 129, JANUARY 2007 Transactions of the ASME

Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 9 Diagnosis of flow structure induced by design case B geometry at design point in terms
of contours of relative Mach number at 3%, 50%, and 100% „rotor blade tip… span. IGV and exit
region are not shown for clarity.

in design case A. The convergence rate in camber during the de- level of convergence was ultimately reached as in the isolated
sign calculation was lower than in the isolated designs. This was designs. The rotor geometries obtained here are denoted as R1B
expected, because the blades have to adapt to changing inlet and and R2B.
exit conditions during the design calculation. However, the same Results of the subsequent design point multistage analysis with

Journal of Turbomachinery JANUARY 2007, Vol. 129 / 115

Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 10 Multistage analysis of design speed throttling „varying backpressure, pb… for case A in terms of
spanwise profiles of axisymmetric-averaged rotor exit cumulative total pressure

R1B and R2B are shown in Fig. 9. From the contours of relative
Mach number in Fig. 9, it can be seen that both rotors now pro-
duce the implied flow structure of the design-intent pressure load-
ing. Both rotors have passage shocks in their intended locations.
In particular, the passage shock positions in both R1B and R2B
are much closer to those shown in Fig. 7 than the case of R1A and
R2A shown in Fig. 8. The flow field near the hub of stator 2 has
improved somewhat relative to that of the isolated design 共Fig. 8
versus Fig. 9兲.
Compressor Performance at Design Speed. Next, a design-
speed multistage analysis of the redesigned rotors was performed
at various backpressures from design to near stall 共pb
= 4.0– 4.15兲. Note that the original design shown in Fig. 4 has
very limited range; rotor 1 shock spills at a backpressure of
around pb = 4.05. Results for case A 共isolated blade design兲 and
case B 共multistage design兲 are shown in Figs. 10 and 11, respec-
tively. Indicated, are plots of spanwise profiles of axisymmetric-
averaged 共mass weighted兲 rotor exit total pressure. It can be seen
that case B has a superior throttling characteristic relative to that
of case A. In particular, with respect to the second rotor, R2A is
hub weak as the machine is throttled 共Fig. 10兲, while the spanwise
total pressure profile of R2B is relatively uniform 共Fig. 11兲. With
respect to the first rotor, R1B is completely insensitive to the
compressor backpressure variations 共Fig. 11兲, while R1A is insen-
sitive to compressor throttling up to pb = 4.10.
Figure 12 shows comparison of the Mach number contours at
the midspan station between case A 共plots on left side兲 and case B
共plots on right side兲. Inspection of these Mach number contours
shows that above the backpressure value of pb = 4.10, the passage
shock in R2A spills while it is still well inside the blade passage in
R2B. It is also observed that the passage shock in R1B is much
weaker than in R1A over the throttling range.
The differences in design-speed performance characteristic be-
tween case A and case B are further illustrated in Fig. 13 in terms
of rotor total pressure ratio and adiabatic efficiency variations
with compressor backpressure. This performance variation is
strongly dictated by the passage area distribution of each rotor.
Both rotors have a certain amount of internal contraction resulting
from the prescribed pressure-loading distribution, which is in-
tended to reduce the passage shock strength at design speed with
possible negative effects on part-speed performance. It is clear
that R1B remains choked throughout the throttling range as is
evident by the nearly constant performance characteristic with
varying backpressure. The rates of change of pressure ratio and

Fig. 11 Multistage analysis of design speed throttling for case B in terms of spanwise profiles of
axisymmetric-averaged rotor exit cumulative total pressure

116 / Vol. 129, JANUARY 2007 Transactions of the ASME

Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 12 Multistage analysis of design speed throttling for case A „left… and case B „right… in
terms of relative Mach number contour at midspan for different backpressures
efficiency with backpressure for the two cases are very different
indicating a superior design point blade row matching potential
for case B. Assessment of the aero-stability of the 2.5 stage is
Fig. 13 Multistage analysis of design speed throttling for
cases A and B. Shown are throttling characteristics of the re-
spective rotors 1 and rotors 2 within the 2.5 stage.
Journal of Turbomachinery
Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
computationally difficult and is subject to many uncertainties that
are beyond the scope of this analysis. Heuristically, it would seem
that R2B has a lager peak pressure rise potential than R2A, which
does not necessarily imply a more favorable compressor in terms
of ultimate throttling range. Finally, Fig. 14 shows that case B also
exhibits a slight increase in adiabatic efficiency over the applied
design-speed throttling range when compared to case A, on the
order of 0.3%. Figure 14 indicates that for both case A and case B,
the design requirement of an overall compression ratio of 4.65:1
at a backpressure of 4.0 is nearly satisfied.
In summary, this exercise shows that by using the newly imple-
mented multistage inverse method, the resulting rotor blade de-
signs 共design case B兲 produce pressure-loading distributions
closer to design-intent than when using the isolated blade row
inverse method 共design case A兲. The rotor blades designed using
the multistage method also exhibit a more desirable design-speed
throttling characteristic. Whether or not this leads to a more fa-
vorable aerostability characteristic requires further careful inves-
tigation.
Conclusions
An existing three-dimensional inverse blading design code,
INV3D, has been successfully extended to handle multistage blad-
ing design and analysis via a steady averaging-plane blade row
coupling scheme. The extended code has been successfully ap-
plied to redesign exercises involving a highly loaded 2.5-stage
axial compressor.
It has been shown that multiple blades can be redesigned simul-
taneously in their mutually interacting environments with a mod-
est increase in computational cost. Furthermore, the viability of
JANUARY 2007, Vol. 129 / 117
 Fig. 14 Multistage analysis of design speed throttling for cases A and B. Shown are the
throttling characteristics of the complete 2.5 stages.
inverse blading design for the purpose of enhanced design point
aeromatching of multistage compressors has been demonstrated.
This has the potential to greatly facilitate the process of blade row
matching during design iterations, and is a significant improve-
ment over existing approaches used to address design point aero-
dynamic matching.
Many more issues need to be addressed before this inverse
design scheme can be successfully deployed in the future design
system proposed herein. Noteworthy, is the need for a pressure-
loading regulation scheme that automatically adjusts the pressure-
loading distribution and magnitude in order to meet the required
axisymmetric-averaged design intent static pressure field. It is
hoped that the present investigation will serve to motivate further
research.
Acknowledgment
This work was partially funded by NASA Glenn Research Cen-
ter for whose support the authors are grateful.
References
关1兴 Adamczyk, J. J., 2000, “Aerodynamic Analysis of Multistage Turbomachinery
Flows in Support of Aerodynamic Design,” ASME J. Turbomach., 122, pp.
189–217.
关2兴 Gallimore, S. J., 1999, “Axial Flow Compressor Design,” Developments in
Turbomachinery Design, J. Denton, ed., American Institution of Mechanical
Engineers.
关3兴 Marble, F. E., 1964, “Three-Dimensional Flow in Turbomachines,” High
Speed Aerodynamics and Jet Propulsion, Vol. X, Aerodynamics of Turbines and
Compressors, W. R. Hawthorne, ed., Princeton University Press, Princeton,
NJ, pp. 83–165.
关4兴 Dang, T. Q., 1995, ‘ ‘Inverse Method for Turbomachine Blades Using Shock-
Capturing Techniques,” AIAA Paper No. 95-2465.
关5兴 Qiu, X., 1999, “Improved Algorithm for Three-Dimensional Inverse Method,”
Ph.D. dissertation, Department of Mechanical Engineering, Syracuse Univer-
sity, Syracuse, NY.
关6兴 Damle, S., Dang, T., Stringham, J., and Razinsky, E., 1999, “Practical Use of
a 3D Viscous Inverse Method for the Design of Compressor Blade,” ASME J.
Turbomach., 121共2兲, pp. 321–325.
118 / Vol. 129, JANUARY 2007
Downloaded 08 Mar 2009 to 194.225.236.227. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
关7兴 Qiu, X., and Dang, T., 2000, “3D Inverse Method for Turbomachine Blading
With Splitter Blades,” ASME Paper No. 2000-GT-526.
关8兴 Dang, T., Damle, S., and Qiu, X., 2000, “Euler-Based Inverse Method for
Turbomachine Blades: Part II—Three Dimensions,” AIAA J., 38共11兲, pp.
2007–2013.
关9兴 Medd, A. J., 2002, “Enhanced Inverse Design Code and Development of De-
sign Strategies for Transonic Compressor Blading,” Ph.D. dissertation, Depart-
ment of Mechanical Engineering, Syracuse University, Syracuse, NY.
关10兴 Jameson, A., Schmidt, W., and Turkel, E., 1981, “Numerical Solution of the
Euler Equations by Finite Volume Methods Using Runge-Kutta Time-Stepping
Schemes,” AIAA Paper No. 81-1259.
关11兴 Denton, J. D., 1992, “The Calculation of Three Dimensional Viscous Flow
Through Multistage Turbomachines,” ASME J. Turbomach., 114, pp. 18–26.
关12兴 Wadia, A. R., and Beacher, B. F., 1990, “Three-Dimensional Relief in Turbo-
machinery Blading,” ASME J. Turbomach., 112, pp. 587–598.
关13兴 Reid, L., and Moore, R. D., 1978, “Design and Overall Performance of Four
Highly-Loaded, High-Speed Inlet Stages for an Advanced, High-Pressure-
Ratio Core Compressor,” NASA TP-1337.
关14兴 Reid, L., and Moore, R. D., 1980, “Experimental Study of Low Aspect Ratio
Compressor Blading,” ASME Paper No. 80-GT-6.
关15兴 Hall, E. J., and Delaney, R. A., 1995, “Investigation of Advanced Counterro-
tation Blade Configuration Concepts for High Speed Turboprop Systems: Task
VII-ADPAC User’s Manual,” NASA CR-195472.
关16兴 Dawes, W. N., 1992, “Toward Improved Throughflow Capability: The Use of
Three-Dimensional Viscous Flow Solvers in a Multistage Environment,”
ASME J. Turbomach., 114, pp. 8–17.
关17兴 Chima, R. V., 1998, “Calculation of Multistage Turbomachinery Using Steady
Characteristic Boundary Conditions,” NASA Technical Memorandum, No.
206613.
关18兴 Giles, M. B., 1990, “Nonreflecting Boundary Conditions for Euler Equation
Calculations,” AIAA J., 28, pp. 2050–2058.
关19兴 Gropp, W., Lusk, E., Doss, N., and Skjellum, A., 1996, “A High-Performance,
Portable Implementation of the MPI Message Passing Interface Standard,”
Parallel Comput., 22, pp. 789–828.
关20兴 Gropp, W. D., and Lusk, E., 1996, “Guide for a Portable Implementation of
MPI,” ANL-96/6, Mathematics and Computer Science Division, Argonne Na-
tional Laboratory.
关21兴 Larosiliere, L. M., Wood, J. R., Hathaway, M. D., Medd, A. J., and Dang, T.
Q., 2002, “Aerodynamic Design Study of an Advanced Multistage Axial Com-
pressor,” NASA TP-211568.
关22兴 Medd, A. J., Dang, T. Q., and Larosiliere, L. M., 2003, “3D Inverse Design
Loading Strategy for Transonic Axial Compressor Blading,” ASME Paper No.
2003-GT-38501.
Transactions of the ASME